6
14

Super-polynomial accuracy of multidimensional randomized nets using the median-of-means

Abstract

We study approximate integration of a function ff over [0,1]s[0,1]^s based on taking the median of 2r12r-1 integral estimates derived from independently randomized (t,m,s)(t,m,s)-nets in base 22. The nets are randomized by Matousek's random linear scramble with a digital shift. If ff is analytic over [0,1]s[0,1]^s, then the probability that any one randomized net's estimate has an error larger than 2cm2/s2^{-cm^2/s} times a quantity depending on ff is O(1/m)O(1/\sqrt{m}) for any c<3log(2)/π20.21c<3\log(2)/\pi^2\approx 0.21. As a result the median of the distribution of these scrambled nets has an error that is O(nclog(n)/s)O(n^{-c\log(n)/s}) for n=2mn=2^m function evaluations. The sample median of 2r12r-1 independent draws attains this rate too, so long as r/m2r/m^2 is bounded away from zero as mm\to\infty. We include results for finite precision estimates and some non-asymptotic comparisons to taking the mean of 2r12r-1 independent draws.

View on arXiv
Comments on this paper